Improving box office projections through sentiment analysis: Insights from regularization-based forecast combinations

Abstract

The increasing reliance on box office revenue as a real activity tracker necessitates a deeper understanding of how sentiment affects this micro-level indicator. Although previous research has attempted to model the nonlinear relationship between sentiment and box office performance, the issue of model uncertainty remains unresolved. In this study, we analyze data from the North American film market between October 1, 2010, and June 30, 2013, to examine the potential benefits of adopting forecast combination approaches for predicting box office performance using sentiment data. Our results indicate that regularization-based forecast combination significantly improves predictions compared to individual forecasters and alternative combination methods. Furthermore, we propose a novel predictor ranking approach utilizing bootstrap resampling, which reinforces the significance of all five sentiment measures as top predictors, thereby offering valuable insights into the correlation between sentiment and box office revenue.